Math626123-03-24

Review.

im. Suppose that X., X2,.. are

independent with mean o.

E,Var(Xn)sr,
E

If then
En converges with
prob.1

· The
following result, called the three-series the, provides necessary
and sufficientconditions
individual
for the
convergence of
5 X in

terms of the distribution

of
Xn.

letX" Xn truncated at
For 220, =

Xn11nIss) be

-4.6 (Kolmogrow's three-series ihm). Let X, X2,.; be

independent rv's. Let cco. Then in order that

Xn
and
converges a.s., it is
necessary sufficient that

",P)(Xnkc) so,

(2)
EX?" so,
(3)
-Var(Xn") si.
 that the three
#
Sufficiency: Suppose series
converge

Write EX.
mn=
Bythan 4.4,

-(X -

m) converges
ans.

Since in converges,
it follows
E
that I X"converges a.s.

Since =,P(IXn)>c) co,

by the
firstBorel-Cantelli lemma,
P((X)2 i.0) 0, =

which that
implies
P) XnEX"i.0.) = 0.

Hence it follows that

X converges
a.s.

of of necessity.

Suppose that zXn converges

as.
Fix c20.

Since Xnto as, follows that

it IX converges are.

the second Borel Cantelli lemma,

and
by
P(1Xn1 e) so.

In what follows we show that

Var(X2") so
Waite MmE E) zXn")
Si Var(
=

Xn").
Assume, on the contrary, that Sati as n+0.

Since X"-E(X?") are

uniformlybounded, bythe ChT,

hi PJo AixnaMn, y)
=Se *Ydz. (**)

Since EXn converges, sn-two also

implies that

+Xn -> 0 a.s.

Hence
p) /* 1 :
a) -> 0 *(
(**

Now (**) and ****) stand in contradiction.Since

PJx Fit-Mn cy, (Ex)

positive for
is all
sufficientlylarger (if x <y), but

then
x-s < sy+ s.(****)

However, (****) can not hold simultaneouslyfor

 say, (x-9, 4+2) =
( -

1,0) and
(x-5,4 3) + =
(1, 2).

Hence we have him so so, that

is, Var(Xn") so

Now the 4.4,

by
E(Xn=EXn converges. A.S.
M=1

Since we
Xn converges, have E, EX
converges.

the
Now we turn to
properties of random walks.

Def. (Random walks).

Let X, X2, ..., be i.i.d. random vectors RP.

in Set

Sn Xit = ...
+

n.
Then Si is called a random walk.

For convenience, we build a new

prob. space
v
Sw
=
=
(w,w,...). Wie RYS
F B(R*)
=

x
. .. x (RY) x ...

D fexx
=
x ...

xM...
 where the law of Xi.
he is

Set Xn(w) Wn.

=

Let A =
$1,2, ...
3 be the set of positive integers.
f. A
in
permutation it:AT-AT is a

map
such that

#(i) E i

for onlyfinitelymany i.

For a
finite permutation it
of AT WE-R,
and
define
(w); Wisil.
=

That is, the coordinates of ware rearranged accordingto it.

Def. An event A I & is called Parable if

P(A) P(TA)
=

for any finite permutation to

In other words.
permutable if the occurrence of
A is A
is not
affected if re-arranging finitelymany of the r.u.s
 the collection 8-field.
Def of all permutable events is a

is
It called the eme-field and is denoted by E.

Then
Example:Supposed-1.
①(w: Sn(wIEB i.0.)
② (w: limup. I
are
permutable. This is because
for each finite permutation it,

Sn(w) =
SnCic) for largen.

Fact: All events the tail

in o-field are
permutable.
To see it, notice that if8(Xnt, Xutz, ...), then the
At

occurrence
of A is
unaffected by permutation of Xr..
a
A
 The
followingresult generalizes Kolmogrow's 0-1 haw:

Thi 4.5)Hewitt Savage - 0-1 law).

id At E. Then
If X, X2, ...,
are and

P(A) 0
= or 1.

Pf. Let At 9.
The show that A
idea is to is
independent itself.

Since 8(X, X2,

At ...), there exists Ant8(X, X2, ..., Xn)
suchthat

P(An A) 0,

Here AOB =
(AB) U
(BA) is the symmetric difference.

Notice that An can be written as

An= (w: (wr, wn, ..., wn) Br

-

for some Br.

Now define a
permutation it by

if jan

I
i() jh
+
15

j
=

j
-
n
if n+1 < =
2n

j if jx 2n.

Then P po i
=
 p(AndAl =
PSW: wE
AndAY
PSW:πW-AndA 3
=

PSW:WE i'An iAl

=
c

rAn i'A=
PSW:WE
=

A), since A.

Write AnAn.

Then An=SW: (Wnt, ..., Win) BrJ.

e

So An and Anare
independent.
Now P(AndA) to

P(AnA) ->o

Since An An =
(AnA)U (An Al
So

P(AndAn to.

Notice that

0 <
P(An)-P(AndAn

=P(AnwAn) =

P(AndAn
P(AncAn)
=

to
 So
p(AndAn) ->

P(A).

But P(AntAnD P(An) P(Ans = ->

P(A)2.
It
follows that

P(A) P(A)2.=

1.

As an
application, we have the following.

im
4.8: For a random walk on
IR, there are
only
4 possible cases, one
ofwhich has
probability1.
(i) Sn 0
=

for all n.

(ii) Sn =>
+ 0.

(iii) Sn--0

Sn Sn
(iv) lining >
limsup 0.
=

- 0 =

Pf. Byihm 4.7.

limsup S c
=

7 1,01
= -

as.

Let Sn= Sn+-X.Since Sn+i-XIhas the same

distribution,

It follows that c = c =

x1.
If c is finite, then X1=0 and (i) occurs.
If I is not finite, then C = +0 or-0.

The
same
analysis applies to the
liming.
#.